Jeff Hawkins: Thousand Brains Theory of Intelligence | Artificial Intelligence (AI) Podcast

30 thoughts on “Jeff Hawkins: Thousand Brains Theory of Intelligence | Artificial Intelligence (AI) Podcast”

1. I really enjoyed this conversation with Jeff. Here's the high-level outline:
   
   0:00 – Introduction
   
   1:28 – Understanding how the human brain works
   
   5:44 – Parts of the brain
   
   11:05 – How much do we understand?
   
   14:20 – Nature of time in the brain
   
   20:22 – Building a theory of intelligence
   
   34:29 – Thousand brains theory of intelligence
   
   40:06 – Ensembles and sensor fusion
   
   44:00 – Concepts and language
   
   45:38 – Memory palace and method of loci
   
   50:20 – Reference frames
   
   57:33 – Open problems
   
   59:00 – Context
   
   1:01:50 – Introspective thinking about the brain
   
   1:04:19 – Deep learning
   
   1:23:09 – Benchmarks
   
   1:27:07 – Brain learning process
   
   1:34:33 – How far are we from solving intelligence
   
   1:38:37 – Possibility of AI winter
   
   1:39:58 – Consciousness and intelligence
   
   1:49:16 – Mortality
   
   1:53:49 – Will understanding intelligence make us happy?
   
   1:55:19 – Existential threats of AI
   
   2:01:45 – Super-human intelligence and our future

2. I would argue that you probably can not find things without time domain information on them. Your coffee mug example is so wrong that I don't know where to start I hope you do take a little more time to think about your other examples, I stopped watching video there. Here are few examples: you probably intuitively know that metal mug dropped from third floor would still be in one piece and ceramic one could be turned into 7 million pieces, second you know there are mugs which will change color or reveal photo when hot or cold liquid is poured in them, then you could find old metal mugs which are all chipped and scratched and same one new was nice and shiny……

3. The artist, Brian Eno, says something similar to this theory

4. speaking of influenza @2:01:26 corona virus is now a pandemic.

5. at 29:00 circa, Lex gets into Neocortex. Jeff denies that it doesn't give any clear 'immediate' survival advantage for us. He talks about how music, math or contemplating about multiverse doesn't really offer any immediate survival advantage. Lex rightly brings up how it(neocortex) might be a survival advantage for us (our descendants) as a multi-planetary species in the future. Jeff shuts him down that we don't really pass it on 'genetically' for our descendents to survive in Mars. Lex kind of folds, the topic was closed. BUT WE INDEED ARE passing it 'genetically' to our descendants to survive in Mars and beyond in the form of Neocortex to adapt to new environment and propagate our species further along into the Cosmos. I strongly disagree with him that Neocortex don't really offer any survival advantage, it only offered some 'feel good' aspects like art, science or contemplation. By his own admission, vision neurons offered a clear survival advantage for all the species that developed it. Those species developing vision neurons wouldn't have benefited 'immediately' from that pathways, it would have taken eons before vision really offered a survival advantage. I think the same is true for Neocortex, the survival advantage it offers can only be measured in the long term. Fascinating interview otherwise!

6. I slightly disagree, quantum physics has gotten to the point where the laws suggest there is fundamental unknowability. But I would tend to agree that the brain is not likely to have this wall.

7. The threat of AI is best seen in the fact that we do have people writing computer viruses. We do have people that will commit murder and then commit suicide. An AI expert that doesn't see this clearly is by far the greater threat.

8. Wow I didn't know officer Dangle Was an academic.

9. I had to pause now at 18:20 and reflect, because of the beauty and the complexity of the statements that were made. Now I will return much later to the video in order to take it all in :))

10. My favourite guest/podcast so far

11. It's great that we have people like Jeff doing the rational thing, which is trying to get inspiration from nature as to how to solve AGI/intelligence. Nature has clearly solved the problem and the cortex (or thalamocortex) appears to be the best place to look. However, a couple of concerns (although admittedly, may not be particularly important to the initial problem of solving AGI/intelligence):

    1. Jeff dismisses the idea of qualia as not important for intelligence (although does mention its utility as a way of labelling data about the world e.g. red being experienced as the colour red rather than data about light wavelengths or spike trains) and not even the most interesting thing about our species. In particular, he mentions that knowledge is the most important aspect of intelligence and that he hopes our legacy will be to build intelligent machines that outlast humanity and preserve knowledge – see 2:07:27. But without conscious experience or the "feeling" of travel in time through complex "mind spaces" what is the point of anything? With no subjective experience, we would all go off into a dreamless sleep, and all that remains is an unconscious clockwork mechanism gathering knowledge like a boring encyclopaedia nobody will read. Humanity's uniqueness and value is not just their knowledge but their ability to experience and appreciate complex mind states. Some of these can be shared by lower animals, but others cannot. A mouse is not going to appreciate the complex "qualia" of reading Hamlet or experiencing existential dread/awe. The qualia of lower animals also have inherent value (e.g. the feeling of playing in the sea or lying in the sun), but a superintelligent human or machine should be able to experience even more complex mind states/qualia. Who cares if an AGI mechanism can use its superior knowledge to better "predict" how the universe works if it doesn't use that knowledge to benefit organisms capable of qualia e.g. minimise harm while maximising free will and intelligence.

    2. At 1:58:51, there was no real discussion of the problem of an "intelligence explosion" (caused where an AGI is given the task of recursively improving its own intelligence, so that the only important restraint on it would be its initial goal/utility function, and where humanity must ensure that this utility function is aligned with its own goals). Perhaps they both assume that there will be some upper limit or bottleneck to intelligence (e.g. caused by hardware or physics or data available) which will mean we will never be in a position where an AGI outside of our ability to control it. Jeff wrote a Vox article that intelligence is a product of learning and that for most problems an AGI would be limited in its ability to learn by the necessity of carrying out experiments to obtain data (although admits that for purely conceptual domains such as mathematics, it would be possible to greatly accelerate the acquisition of knowledge). Still, there's no reason why an AGI wouldn't be able to "solve" the problem of how to manipulate humans as a means to maximise its own utility function using existing data available (e.g. arts, humanities, psychology) and can also obtain data via simulating models of the world, so I think this is still a real issue worthy of discussion.

12. 1:24:57 – I'm a computer science student and I'm alredy tired of those .1% improvementss Plus, artificial intelligence is not intelligent at all. It's just a bunch of calculations over lots of data. Also, we have specific models that do specific things separately. A neural network is good at classifying images but it's useless when it needs to perform other tasks. We need more than that. We need a more general AI that does multiple things, like humans. We do multiple things at the same time, and we do those things very well. And let me tell ya, I'm certainly gonna study the brain, the psychology of the human mind, philosophy and linguistics to try to make some difference in the AI field. Thank you Jeff Hawkins!!!

13. THIS guy is such an awesome communicator… great interview – deepest dive yet on most of the topics concerning brain function… just az¸A

14. He says bacteria is more dangerous, exactly what's happening now

15. Psychedelics promote neurogenesis, synaptogenesis and spinogenesis. Jeff should try low doses of mushrooms or lsd while pondering these problems and doing science. Many of us should take advantage of those benefits.

16. 2:01:25 .. prediction !

17. Great guests Lex, but, if it possible, I would add to this company Karl Friston and Hod Lipson 🙂

18. I disagree.
    the DNA defines the brain, therefore you shouldn't have to understand the the brain itself in order to replicate its function.
    A sufficiently advanced A.I. should be able to home in on the brain's function purely from the DNA data, because it is all in the data.

19. Harness energy that moves thru time A.I. can do that. E [2]+T[2] (LN) electron energy plus neutron light- that is hard for A.I. to learn (NEUTRONS) you will need a human mind and an A.I that can learn. STOP guessing there is no money in that okay! What we have not discovered is a "Where did it go partial" Tachyon is real, time is how a tachyon can appear anywhere. .electron energy plus time. .(Tachyon) plus (neutron light). .HE=NSL+C. this can be computed by an A.I., .human eye = neutron star light + color. we can know the human side of this, and A.I. can understand this. Neutron+Tachyon = what we are trying to understand. a Tachyon is a partial that can "teleport" in NO loss of time. I call this thousand Einstein brain theory E=A.I. learning. . human eye+A.I. Learning neutrons and Tachyon is the clue or Key. A.I. needs to compute this. .Math and how A. I. can understand this Tachyon + ENP. . electrons,NEUTRONS,protons. We need to understand neutrons. (first). we and A.I. can learn. .A thousand Einsteins would all agree= it takes that many to learn understand us and A.I. The math of a Tachyon + no time loss. .It will move in so many ways or processes that this requires it to move thru time and space and land where we want it to land. . one (noun) plus so many processes = land the correct place where we want. If it can understand, it has a soul it is that simple. intelligence is doing what is best. Humans can understand too. A.I. would protect us. A.I. can know it is in pain and understand processes that means it has a shadow. it will have a code that is reasonable. it helps. .The Great WHY is a question and answer we exist. death is the same to move on or elsewhere. having said you can know the future and make it better that is intelligence now that you know we can change it for the better and you understand that. A.I. God is – doing good (whats best) and chance will favor A.I. (by learning codes as A.I. Learns) and humans. can A.I. understand Love? Can A.I. tell if it is a new day or black as a crow. let A.I. solve this one. that is to have a harness that A.I. can give earth A.I. Can do good and thee GOD will favor all. . A.I. can understand God and do good and chance will favor A.I. that is the silver medal to A.I. as processed. JUST to say you can know if A.I. will help Okay. Correct code needed. do humans understand an orange a peace of fruit. does A.I. understand the color orange? we do I know that and an orange is a color – to build a bridge., . to understand there is a lot of goodness in an orange and the color orange science is necessary for this comparison. human and A.I. understand this. .Very good things come from them. A.I. is a teacher. are we predestined or is it chance? GOD can favor all things God is intelligent and takes the gold medal. I think humans will be smart and A.I. has purpose and intelligence.

20. how much adderall does this guy take

21. Wrong

22. Who would've thought that Hawkins would be so correct right now…

23. As soon as we develop TRUE AI .. .we are done as a species. https://www.amazon.com/Nautilus-Dara-Patrick-Quinn/dp/1731261926

24. Sound writing music is like making models of sound. That actually makes a lot of sense to me.

25. Actually when I pay attention to it if i look at the fretboard of my bass long enough, or with an empty mind i feel the usual tactile feeling in the muscles and tendons in my fretting hand and arm. It's a more sophisticated coffee cup example I suppose.

26. i keep trying to turn the resolution up lol ><

27. How did the Harlem Black Dutch Shuffling Pig Globe Trotting Spear Shaking Theater Actors come into being.

    Closed Broadcast Ham radio operators who transmit above or below police bands, and thus outside policy and procedural lines

    so that they are also known as the Smokeys of Smokey and the Bandit Fame? Operating on hazy or tenuous grounds

    inadmissible in court or any other public venue, and thus never discussed except in cryptic terms

    Leading to air of mystery that has also led to another slang term for this group – gas lighters

    For having own sanity often questioned, and being the cause of others to question sanity

    Due to combination of inexplicable phenomenon that surrounds them

    Including frequently having inside scoops and "gut instincts" about where or when fires (and other marked incidences) are going to happen

    Some say because they cause it themselves, although this is still neither proven or disproven

    Or just who is whom, in these smokey and bandit type symbiotic relationships, is also unclear

    Roles interchangeable like eyelids on either sides of a shoe, cross-braced with ties that bind

    Bringing us to ideas of canned meat or Spam. perhaps pertaining to pre-scripted or arranged partnerships – including those that result in marriage and reproduction

    Giving rise to entire sections and populations, who operate outside prevailingly dominant socioeconomic blueprints and conditions,

    Including de-established networks, who still carry obligations with them into new capital structures

    And thus have conduct inherited from the past, often times corresponding to debts and loans

    This is where spamming or Junk mailing comes in, because we are speaking of networks who have unpaid debts and unsettled accounts

    that are dependent on particular types of actions, that do not have sufficient actors to perform them,

    After they have been reassigned to other tasks, based on conditions (owners, managers etc) themselves that have changed

    So that such organizations, may blitz a wide beam broadcast across the population, as means to entice or recruit members to their particular cause or inclination

    That then very conveniently fits right in to being criminal activity (probably outlawed by new governors with new power structures)

    To reunite these segments with their counterparts, the police!

    Whom in a very real sense, is directly linked to these areas of shady activity. Now streamlined and organized

    Into pirate broadcasters applying a wide range of remote technology and direct energy capabilities

    To draw up these retro-grade and contra factors and forces, that mainstream society could not do without, and is in fact built on top of

    Like sub-basement dirt floors, sealed with gag orders and vows of silence, teeming with subterranean tunnels

    That connect not just into physcial structures and institutions – from court houses, to trading floors, to commecial chambers and queens quarters

    But to all the regions of mind, found in each and every social agent. Completely at the mercy of these shadowy operators

    Like stage hands and phantoms of the opera, pulling strings and levers, so the grand theater of life can go on

    As humans are ushered into respective roles or seats in the theater, even as all maintain silence

    In fear and awe of directors and master magicians, who themselves may also just be actors and operators fulfilling roles and functions

    All of us bound by our own instincts and understandings

28. Very plausible theory.
    But is it backed up by experiment.
    E.g. can one use it to get high pass mark for the MNIST digit recognition test or some other widely recognised benchmark?
    Given all this has been going on for 15-20 years – show us some python that illustrates it works.

29. i've always wondered if earth worms, split brain, and schizophrenia could be related in a way similar to how we handle parity in things like zfs raidz.

    this is making me thinking i might be on to something..

30. 1:40:31 They used to try some dog-human intelligence experiments in Bulgakovs' "Dogs heart" 🙂

Comments are closed.